PeerDAS sampling clarifications #3782
Conversation
- describe sample selection - describe sample queries Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
Clarify that what matters is the false positive threshold, allowing different sampling strategies as protocol compliant behavior. Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
While the technique was introduced as LossyDAS, we don't need the name in the specification itself. Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
add LossyDAS sample count generation helper function Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
Importing scipy is not preferred. This is a self-contained version. Eventually an import of math and use of math.comb makes it simpler. Solving other formatting issues as well. Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
|
types, type conversions, and tests and look good to me, thanks! |
Co-authored-by: Pop Chunhapanya <haxx.pop@gmail.com>
Here we assume uniform random selection without replacement. If other methods are used, the target false positive threshold is the main rule to follow. Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
Signed-off-by: Csaba Kiraly <csaba.kiraly@gmail.com>
|
There is no unique naming convention for these. We follow the naming of the direct replacement scipy.stats.hypergeom.cdf function. |
|
|
||
| ```python | ||
| def get_extended_sample_count(allowed_failures: uint64) -> uint64: | ||
| assert 0 <= allowed_failures <= NUMBER_OF_COLUMNS // 2 |
There was a problem hiding this comment.
Nit: Never seen a function docstring below the asserts. Do we do this elsewhere in the specs?
There was a problem hiding this comment.
we did in phase0 process_attestation:
consensus-specs/specs/phase0/beacon-chain.md
Line 1824 in 258c2c9
There was a problem hiding this comment.
I've moved it, although a bit late. Anyway, if needed, here it is 195ec21
This is to clarify sample selection and sample query based on previous discussions.
Note: I think some of SHOULD might be elevated to a MUST here. I was not doing it as the baseline used SHOULD, but I think some of this behaviour should be mandated, even if we can't enforce it.